Storm-Time Characteristics of Ionospheric Model (MSAP) Based on Multi-Algorithm Fusion

Abstract

Geomagnetic storms induce ionospheric disturbances, affecting short-wave radio communication systems. Accurate ionospheric total electron content (TEC) prediction is vital for accurately describing the short-wave radio environment of the ionosphere. We use the Multi-Step Auxiliary Prediction (MSAP) model, a deep learning algorithm, to forecast TEC during geomagnetic storms. The MSAP model integrates Bi-LSTM networks, an auxiliary model, and convolutional processes for spatiotemporal modeling. Our validation shows the MSAP model outperforms the IRI-2016 model in predicting global TEC for the next 6 days in the test set. We assess its performance during 116 geomagnetic storm events, considering storm intensity, solar activity, month, and Universal Time (UT). The MSAP model exhibits a weak correlation with storm intensity and a strong correlation with solar activity. Monthly variation displays similar strong correlations in root mean square error (RMSE) and R² for both models. For UT variation, the other metrics exhibit a weak correlation with the number of Global Navigation Satellite System stations, except for the RMSE of the MSAP and IRI-2016 models.